Parameters50.3M
GFLOPs151.6
Input Size640px
Best mAP61.4%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
DINOv3-distilled ViT
Neck
HybridEncoder
Head
DEIM
Benchmark Results
Performance on COCO val2017 across different hardware configurations
| Hardware | Runtime | mAP@50-95 | FPS | Latency | VRAM |
|---|---|---|---|---|---|
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 61.3% | 2.3 | 444.9ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 61.3% | 2.7 | 370.1ms | 294 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 60.8% | 5.4 | 184.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 61.3% | 5.2 | 194.2ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 61.4% | 27.1 | 36.9ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 61.3% | 17.4 | 57.6ms | 292 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 60.8% | 37.9 | 26.4ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 61.3% | 38.1 | 26.2ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
17.0ms
350.1ms
3.0ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreDEIMv2x.pt")
# Run inference
result = model("image.jpg", conf=0.25, iou=0.45)
# Process results
print(f"Found {len(result)} objects")
print(result.boxes.xyxy) # bounding boxes (N, 4)
print(result.boxes.conf) # confidence scores (N,)
print(result.boxes.cls) # class IDs (N,)detrnms-freePaper: 57.8% mAP